fix(shell): handle bashlex parsing errors for bash builtins like 'time' (#799)

* fix(shell): handle bashlex parsing errors for bash builtins like 'time'

Adds error handling to split_commands() to gracefully fall back when
bashlex can't parse certain bash reserved words and builtins (e.g., 'time').

When bashlex.parse() fails, the function now:
- Logs a warning about the parsing failure
- Returns the script as a single command

This follows the same fallback pattern used for shlex parsing
elsewhere in the file.

Fixes #798

* test(shell): add test and documentation for bash reserved words fallback

Added comprehensive test for split_commands() handling of bash reserved words
that bashlex cannot parse (like 'time'). Improved documentation to explain
why bashlex fails on reserved words and why fallback behavior is correct.

This addresses feedback on PR #799 showing that bashlex (Python port of GNU
bash parser) cannot handle bash reserved words and the fallback approach is
the correct solution.

Co-authored-by: Bob <bob@superuserlabs.org>

Author: TimeToBuildBob

gptme/tools/shell.py

@@ -1065,7 +1065,25 @@ def split_commands(script: str) -> list[str]:
     # Preprocess script to handle quoted heredoc delimiters that bashlex can't parse
     processed_script = _preprocess_quoted_heredocs(script)
 
-    parts = bashlex.parse(processed_script)
+    try:
+        parts = bashlex.parse(processed_script)
+    except Exception as e:
+        # Fall back to treating script as single command if bashlex can't parse it
+        # bashlex (a Python port of GNU bash parser) cannot handle bash reserved words
+        # like 'time', 'coproc', etc. These are special keywords in bash that have
+        # different parsing rules. When bashlex encounters them, it raises an exception:
+        # "type = {time command}, token = {time}"
+        #
+        # We fall back to treating the entire script as a single command, which is
+        # correct behavior since reserved words typically apply to entire command pipelines.
+        # This preserves the user's exact command and allows bash to handle the reserved
+        # word correctly during execution.
+        logger.warning(
+            f"bashlex cannot parse bash reserved words (e.g., 'time'). "
+            f"Treating script as single command. Error: {e}"
+        )
+        return [script]
+
     commands = []
     for part in parts:
         if part.kind == "command":

tests/test_tools_shell.py

@@ -247,6 +247,32 @@ def test_split_commands_heredoc_quoted_with_spaces():
     assert '<<"EOF"' in commands[0]
 
 
+def test_split_commands_bash_reserved_words():
+    """Test that split_commands handles bash reserved words that bashlex can't parse.
+
+    bashlex cannot parse bash reserved words like 'time' and will raise an exception.
+    In these cases, split_commands should gracefully fall back to treating the
+    script as a single command.
+    """
+    # Test 'time' reserved word
+    script_time = "time ls -la"
+    commands = split_commands(script_time)
+    assert len(commands) == 1
+    assert commands[0] == script_time
+
+    # Test 'time' with more complex command
+    script_time_pipeline = "time ls -la | grep test"
+    commands = split_commands(script_time_pipeline)
+    assert len(commands) == 1
+    assert commands[0] == script_time_pipeline
+
+    # Test 'time' with redirection
+    script_time_redirect = "time echo 'test' > output.txt"
+    commands = split_commands(script_time_redirect)
+    assert len(commands) == 1
+    assert commands[0] == script_time_redirect
+
+
 def test_function(shell):
     script = """
 function hello() {